Self-Supervised Contrastive Learning for Potato Leaf Disease Classification

Potato crops, vital to global food security, are often jeopardized by numerous foliar diseases, including late blight, early blight, and bacterial leaf spot. If untreated, these illnesses can lead to substantial yield losses and negatively impact agricultural output. Conventional disease diagnosis methods, which frequently depend on expert manual inspection, are both time-consuming and labor-intensive, as well as prone to human errors. Recent improvements in deep learning-based image classification approaches have shown considerable potential in automating the identification of plant diseases, offering a scalable and more dependable alternative to manual inspections. In this research, we proposed a Self-Supervised Contrastive Learning methodology for the classification of potato leaf diseases using two publicly available potato leaf datasets. Our model attained an accuracy of 99.46% with a loss of 0.013957 and an accuracy of 98.79% with a loss of 0.096844. These results indicate the potential of self-supervised contrastive learning in agricultural disease detection, offering a scalable and efficient solution for real-world disease classification with minimal reliance on labeled data.